fix(services): не балансировать сервис с одним IP
quality / commitlint (push) Skipped
CD / update-wiki (push) Successful in 5s
quality / changes (push) Successful in 8s
quality / docker-check (push) Skipped
quality / web (push) Successful in 55s
quality / api (push) Successful in 52s
CD / quality (push) Successful in 2m4s
CD / publish (push) Successful in 1m51s

KPI и карточка показывают живой OK, а не гистерезис unknown; DNS по-прежнему ждёт повторные успехи.

Co-authored-by: Cursor <[email protected]>
This commit is contained in:
Denozordec
2026-08-20 18:13:06 +07:00
co-authored by Cursor
parent a9a84fabac
commit f1443f1db5
13 changed files with 464 additions and 72 deletions
+42 -17
View File
@@ -51,7 +51,28 @@ describe('toFailoverEvents', () => {
expect(isFailoverEventStatus('unhealthy')).toBe(false)
})
it('Down всегда виден, даже без A-записей', () => {
it('один IP у сервиса — не инцидент пула', () => {
const events = toFailoverEvents(
[
row({
ip: '10.0.0.1',
status: 'down',
consecutive_failures: 9,
last_error: 'fetch failed',
}),
],
[
{
fqdn: 'solo.example.com',
configured: ['10.0.0.1'],
active: [],
},
],
)
expect(events).toEqual([])
})
it('Down без shared pool не событие балансировки', () => {
const events = toFailoverEvents([
row({
ip: '130.49.213.153',
@@ -60,19 +81,7 @@ describe('toFailoverEvents', () => {
last_error: 'fetch failed',
}),
])
expect(events).toEqual([
{
id: '130.49.213.153',
address: '130.49.213.153',
status: 'down',
kind: 'last-resort',
fqdns: [],
consecutiveFailures: 9,
lastFailureReason: 'fetch failed',
lastCheckAt: null,
},
])
expect(failoverEventCopy(events[0]!)).toBe('Down, в A-записях last-resort')
expect(events).toEqual([])
})
it('OK вне пула не инцидент (standby)', () => {
@@ -151,9 +160,7 @@ describe('toFailoverEvents', () => {
},
],
)
expect(events[0]?.kind).toBe('last-resort')
expect(events[0]?.fqdns).toEqual([])
expect(failoverEventCopy(events[0]!)).toBe('Down, в A-записях last-resort')
expect(events).toEqual([])
})
it('down без A-записи на configured FQDN — снятие', () => {
@@ -293,6 +300,24 @@ describe('mergeFailoverHistory', () => {
expect(history[0]?.source).toBe('dns')
})
it('drops probe leave/return when the service has no shared pool', () => {
const history = mergeFailoverHistory(
[],
[
probe({ id: 1, status: 'up', checked_at: '2026-08-20T09:00:00Z' }),
probe({ id: 2, status: 'down', checked_at: '2026-08-20T09:10:00Z' }),
],
[
{
fqdn: 'solo.example.com',
configured: ['130.49.213.153'],
active: ['130.49.213.153'],
},
],
)
expect(history).toEqual([])
})
it('drops dedicated extra-FQDN DNS rows', () => {
const history = mergeFailoverHistory(
[
+19 -7
View File
@@ -44,9 +44,20 @@ export interface FailoverBindingPool {
active: readonly string[]
}
/** Shared pool FQDN — two or more A targets. Dedicated extra-FQDN has one IP. */
/** Unique A targets. Duplicates of the same IP are not a pool. */
export function uniqueIpCount(ips: readonly string[]): number {
return new Set(ips.filter(Boolean)).size
}
/** Shared pool FQDN — two or more unique A targets. */
export function isSharedPoolBinding(binding: FailoverBindingPool): boolean {
return binding.configured.length >= 2
return uniqueIpCount(binding.configured) >= 2
}
export function hasSharedPool(
bindings: readonly FailoverBindingPool[],
): boolean {
return bindings.some(isSharedPoolBinding)
}
/** Инцидент только при down. up / degraded / unknown — не вывод из пула. */
@@ -66,13 +77,14 @@ export function failoverEventCopy(event: FailoverEvent): string {
}
/**
* Текущие Down всегда в панели. Per-FQDN: снята с hostname или last-resort.
* Standby (up / degraded / unknown) — не инцидент.
* Инциденты пула только если у сервиса есть shared FQDN (2+ уникальных IP).
* Один IP — не балансировка и не вывод из пула; Down смотрит health-монитор.
*/
export function toFailoverEvents(
ipHealth: readonly FailoverHealthInput[],
bindings: readonly FailoverBindingPool[] = [],
): FailoverEvent[] {
if (!hasSharedPool(bindings)) return []
return ipHealth.filter((row) => isFailoverEventStatus(row.status)).map((row) => {
const removedFqdns: string[] = []
const lastResortFqdns: string[] = []
@@ -221,9 +233,9 @@ export function mergeFailoverHistory(
...dns
.filter((item) => isPoolFqdn(item.fqdn, bindings))
.map(dnsHistoryItem),
...toIpAliveTransitions(probes).map((transition) =>
probeHistoryItem(transition, bindings),
),
...toIpAliveTransitions(probes)
.filter((transition) => fqdnsForIp(transition.ip, bindings).length > 0)
.map((transition) => probeHistoryItem(transition, bindings)),
]
merged.sort(
(a, b) => eventTime(b.created_at) - eventTime(a.created_at) || a.id.localeCompare(b.id),
+20
View File
@@ -7,6 +7,7 @@ import {
latestHealthByIp,
providerHealthStatuses,
resolveIpDisplayHealth,
resolveServiceDisplayHealth,
worstHealthStatus,
type HealthLogProbe,
} from '@/lib/health-log'
@@ -146,3 +147,22 @@ describe('resolveIpDisplayHealth', () => {
expect(resolveIpDisplayHealth('unknown', undefined)).toBe('unknown')
})
})
describe('resolveServiceDisplayHealth', () => {
it('shows OK when live probes recovered and stored is still unknown', () => {
expect(
resolveServiceDisplayHealth(
'unknown',
[{ ip: '2.59.161.102', status: 'unknown' }],
[
probe({
id: 1,
ip: '2.59.161.102',
status: 'up',
checked_at: '2026-08-20T18:00:00Z',
}),
],
),
).toBe('up')
})
})
+19
View File
@@ -194,3 +194,22 @@ export function resolveIpDisplayHealth(
if (live && live !== 'unknown') return live
return stored ?? live ?? 'unknown'
}
/** Service KPI / card: any-up of per-IP overlay (live probes beat hysteresis). */
export function resolveServiceDisplayHealth(
stored: HealthLogStatus | undefined,
ipHealth: readonly { ip: string; status: string }[],
probes: readonly HealthLogProbe[] = [],
): HealthLogStatus {
const liveByIp = latestHealthByIp(probes)
const statuses =
ipHealth.length > 0
? ipHealth.map((row) =>
resolveIpDisplayHealth(
row.status as HealthLogStatus,
liveByIp.get(row.ip)?.status,
),
)
: [...liveByIp.values()].map((row) => row.status)
return resolveIpDisplayHealth(stored, bestAliveHealthStatus(statuses))
}
+6
View File
@@ -36,6 +36,8 @@ export const serviceGroupsQueryOptions = () =>
}
return parsed.data
},
refetchInterval: 10_000,
staleTime: 5_000,
})
export const servicesQueryOptions = () =>
@@ -101,6 +103,8 @@ export const serviceViewQueryOptions = (id: number) =>
const data = await api.get<unknown>(`/api/v1/services/${id}`)
return serviceViewSchema.parse(data)
},
refetchInterval: 10_000,
staleTime: 5_000,
})
export const serviceHealthLogQueryOptions = (id: number) =>
@@ -110,6 +114,8 @@ export const serviceHealthLogQueryOptions = (id: number) =>
const data = await api.get<unknown>(`/api/v1/services/${id}/health-log`)
return z.object({ items: z.array(healthProbeLogSchema) }).parse(data)
},
refetchInterval: 10_000,
staleTime: 5_000,
})
export const serviceFailoverLogQueryOptions = (id: number) =>
@@ -32,9 +32,11 @@ import {
ServiceHealthMonitor,
} from '@/components/reui-kit'
import { api } from '@/lib/api-client'
import { hasSharedPool } from '@/lib/failover-events'
import {
enabledHealthProviders,
providerHealthStatuses,
resolveServiceDisplayHealth,
} from '@/lib/health-log'
import type { ServiceView, UpdateServiceConfigInput } from '@/lib/schemas'
import {
@@ -115,6 +117,22 @@ function ServiceDetailPage() {
})),
[service?.domains],
)
const uniqueEnabledIps = useMemo(
() =>
(service?.ips ?? []).filter((ip) => service?.ip_enabled[ip] !== false),
[service?.ips, service?.ip_enabled],
)
const displayHealth = useMemo(
() =>
resolveServiceDisplayHealth(
service?.health_status,
service?.ip_health ?? [],
logItems,
),
[service?.health_status, service?.ip_health, logItems],
)
const showPoolPanel =
uniqueEnabledIps.length >= 2 && hasSharedPool(failoverBindings)
const nodes = (nodesQuery.data as OverviewPayload['nodes']) ?? []
const [editOpen, setEditOpen] = useState(false)
@@ -251,7 +269,7 @@ function ServiceDetailPage() {
actions={
<>
<LbModeTile mode={service.lb_mode} />
<HealthCheckBadge status={service.health_status} />
<HealthCheckBadge status={displayHealth} />
<Tooltip>
<TooltipTrigger
render={
@@ -278,20 +296,20 @@ function ServiceDetailPage() {
id: 'status',
icon: <ActivityIcon />,
label: 'Статус',
value: service.health_status === 'up' ? 'OK' : service.health_status,
value: displayHealth === 'up' ? 'OK' : displayHealth,
variant:
service.health_status === 'down'
displayHealth === 'down'
? 'destructive'
: service.health_status === 'degraded'
: displayHealth === 'degraded'
? 'warning'
: 'default',
iconClassName:
service.health_status === 'down'
displayHealth === 'down'
? 'text-destructive'
: service.health_status === 'degraded'
: displayHealth === 'degraded'
? 'text-warning'
: 'text-success',
hint: <HealthCheckBadge status={service.health_status} size="xs" />,
hint: <HealthCheckBadge status={displayHealth} size="xs" />,
},
{
id: 'fqdn',
@@ -305,21 +323,33 @@ function ServiceDetailPage() {
icon: <NetworkIcon />,
label: 'IP',
value: String(service.ips.length),
hint: `${service.active_ips.length} в пуле`,
hint: showPoolPanel
? `${service.active_ips.length} в пуле`
: uniqueEnabledIps.length <= 1
? 'без балансировки'
: service.ips.join(', ') || 'нет',
},
{
id: 'pool',
icon: <ServerIcon />,
label: 'Активный пул',
value: String((overview?.active_addresses ?? service.active_ips).length),
hint: (overview?.active_addresses ?? service.active_ips).join(', ') || 'нет',
label: showPoolPanel ? 'Активный пул' : 'Адреса',
value: String(
(overview?.active_addresses ?? service.active_ips).length,
),
hint:
(overview?.active_addresses ?? service.active_ips).join(', ') ||
'нет',
},
]}
/>
<section
aria-label="Мониторинг"
className="grid min-w-0 items-start gap-2 @3xl:grid-cols-2"
className={
showPoolPanel
? 'grid min-w-0 items-start gap-2 @3xl:grid-cols-2'
: 'grid min-w-0 items-start gap-2'
}
>
<ServiceHealthMonitor
items={logItems}
@@ -327,13 +357,15 @@ function ServiceDetailPage() {
statuses={providerStatuses}
isLoading={logQuery.isLoading}
/>
<ServiceFailoverPanel
lbMode={service.lb_mode}
ipHealth={service.ip_health}
bindings={failoverBindings}
history={failoverHistory}
probes={logItems}
/>
{showPoolPanel ? (
<ServiceFailoverPanel
lbMode={service.lb_mode}
ipHealth={service.ip_health}
bindings={failoverBindings}
history={failoverHistory}
probes={logItems}
/>
) : null}
</section>
{service.ips.length === 0 && service.domains.length === 0 ? (